import tensorflow as tf # from data_provider2 import get_split from tf_utils import start_interactive_session, set_gpu from rnn_models import RNNplusModel import numpy as np set_gpu(5) options = { 'data_root_dir': "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_lrs", # "/home/michaeltrs/Projects/audio23d/data", #"/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_clean", # 'split_name': "devel", # 'devel', 'is_training' : False, 'data_in': 'melf', # mfcc, melf, melf_2d 'max_seq_len': -20, 'use_rmse': False, 'batch_size': 128, # number of examples in queue either for training or inference #'reverse_time': False, #'shuffle': False, #'random_crop': False, #'standardize_inputs_and_labels': True, 'mfcc_num_features': 20, # 20, 'raw_audio_num_features': 533, # 256, 'num_classes': 28, # number of output classes 29 = |a-z, " ", <sos>, <eos>| 'max_out_len_multiplier': 1.0, # max_out_len = max_out_len_multiplier * max_in_len 'mfcc_gaussian_noise_std': 0.0, # 0.05, 'label_gaussian_noise_std':0.0, 'has_encoder': True,
from __future__ import absolute_import from __future__ import division from __future__ import print_function from collections import defaultdict from transformer_model import SelfAttentionEncoder from tf_utils import start_interactive_session, set_gpu set_gpu(7) options = defaultdict( lambda: None, # Set default value to None. # Input params #default_batch_size=2, # Maximum number of tokens per batch of examples. max_length=50, # Maximum number of tokens per example. # Model params initializer_gain=1.0, # Used in trainable variable initialization. vocab_size=28, # Number of tokens defined in the vocabulary file. hidden_size=128, # Model dimension in the hidden layers. num_hidden_layers=3, # Number of layers in the encoder and decoder stacks. num_heads=4, # Number of heads to use in multi-headed attention. filter_size=256, # Inner layer dimension in the feedforward network. # Dropout values (only used when training) layer_postprocess_dropout=0.1, attention_dropout=0.1, relu_dropout=0.1, ################################################## data_root_dir= "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_clean", # enhanced",
import tensorflow as tf # from data_provider2 import get_split from tf_utils import start_interactive_session, set_gpu from mixed_seq_models import CNNRNNModel3 import numpy as np set_gpu(0) options = { 'data_root_dir': "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_dtw_antonio", # "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_dtwN", # "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_lrs", # "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_clean", 'is_training': True, 'data_in': 'melf', # mcc, melf, melf_2d 'split_name': 'train', 'batch_size': 20, # number of examples in queue either for training or inference 'random_crop': False, 'mfcc_num_features': 20, # 20, 'raw_audio_num_features': 533, # 256, 'num_classes': 28, # number of output classes 29 = |a-z, " ", <sos>, <eos>| #'1dcnn_features_dims': [256, 256, 256], 'batch_norm': True, 'has_decoder': True, 'decoder_num_layers': 1, # number of hidden layers in decoder lstm 'residual_decoder': False, # 'decoder_num_hidden': 256, # number of hidden units in decoder lstm
import tensorflow as tf # from data_provider2 import get_split from tf_utils import start_interactive_session, set_gpu from rnn_seq2seq_models import RNNSeq2SeqModel import numpy as np set_gpu(-1) options = { 'data_root_dir': '/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_dtwN', # "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_lrs", # "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_clean", 'is_training' : False, 'data_in': 'melf', # mfcc, melf, melf_2d #'max_seq_len': -20, 'split_name': 'devel', #'data_in': 'mfcc', # mfcc, melf, melf_2d #'use_rmse': False, 'batch_size': 1, # number of examples in queue either for training or inference #'reverse_time': False, #'shuffle': True, 'random_crop': False, #'standardize_inputs_and_labels': True, 'mfcc_num_features': 20, # 20, 'raw_audio_num_features': 533, # 256, 'num_classes': 28, # number of output classes 29 = |a-z, " ", <sos>, <eos>| #'max_out_len_multiplier': 1.0, # max_out_len = max_out_len_multiplier * max_in_len #'mfcc_gaussian_noise_std': 0.0, # 0.05,
from cnn_models import DenseNet1D from tf_utils import start_interactive_session, set_gpu import numpy as np set_gpu(1) options = { 'data_root_dir': "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_lrs", # "/home/michaeltrs/Projects/audio23d/data", # "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_lrs", # , # "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_clean", # enhanced", 'is_training' : True, 'split_name': "train", # ""train", # 'devel', 'data_in': 'mfcc', # mfcc, melf, melf_2d 'max_seq_len': 10, # 'use_rmse': False, 'batch_size': 64, # number of examples in queue either for training or inference # 'reverse_time': False, # 'shuffle': True, 'random_crop': False, # 'standardize_inputs_and_labels': False, 'mfcc_num_features': 20, # 20, 'raw_audio_num_features': 533, # 256, 'num_classes': 28, # number of output classes 29 = |a-z, " ", <sos>, <eos>| # 'max_out_len_multiplier': 1.0, # max_out_len = max_out_len_multiplier * max_in_len 'growth_rate': 20, 'num_layers' : 6, 'final_layer_dim': 64, 'loss_fun': "concordance_cc",
import tensorflow as tf # from data_provider2 import get_split from tf_utils import start_interactive_session, set_gpu from mixed_seq2seq_models import CNNRNNSeq2SeqModel import numpy as np <<<<<<< HEAD set_gpu(2) options = { 'data_root_dir': "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_dtw_antonio", # "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_dtwN", ======= set_gpu(-1) options = { 'data_root_dir': '/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_dtwN', >>>>>>> f798981d5c303deabd8107e5086cbc23a1985d2f # "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_lrs", # "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_clean", 'is_training' : False, <<<<<<< HEAD 'data_in': 'melf', # mcc, melf, melf_2d 'split_name': 'devel', 'batch_size': 1, # number of examples in queue either for training or inference 'random_crop': False, ======= 'data_in': 'mfcc', # mcc, melf, melf_2d #'max_seq_len': -20, 'split_name': 'train', # 'devel',
import tensorflow as tf # from data_provider2 import get_split from tf_utils import start_interactive_session, set_gpu from mixed_seq_models import CNNRNNModel import numpy as np set_gpu(2) options = { 'data_root_dir': "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_dtw_antonio", # "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_dtwN", # "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_lrs", # "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_clean", 'is_training' : True, 'data_in': 'melf', # mcc, melf, melf_2d 'split_name': 'train', 'batch_size': 64, # number of examples in queue either for training or inference 'random_crop': True, 'mfcc_num_features': 20, # 20, 'raw_audio_num_features': 533, # 256, 'num_classes': 28, # number of output classes 29 = |a-z, " ", <sos>, <eos>| '1dcnn_features_dims': [256, 256, 256], 'has_decoder': True, 'decoder_num_layers': 1, # number of hidden layers in decoder lstm 'residual_decoder': True, # 'decoder_num_hidden': 256, # number of hidden units in decoder lstm 'decoder_layer_norm': True, 'decoder_dropout_keep_prob': 1.0,
from cnn_models import CNNModel from tf_utils import start_interactive_session, set_gpu import numpy as np import tensorflow as tf set_gpu(3) options = { 'data_root_dir': "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_dtw_antonio", # "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_dtwN", # "/home/michaeltrs/Projects/audio23d/data", # "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_clean", # "/vol/atlas/homes/pt511/db/audio_to_3d/tf_records_lrs", 'split_name': "devel", # 'devel', 'is_training' : False, 'data_in': 'melf', # mfcc, melf, melf_2d 'batch_size': 1, # number of examples in queue either for training or inference 'random_crop': False, 'mfcc_num_features': 20, # 20, 'raw_audio_num_features': 533, # 256, 'num_classes': 28, # number of output classes 29 = |a-z, " ", <sos>, <eos>| 'has_encoder': True, '1dcnn_features_dims': [256, 256, 256], 'loss_fun': "concordance_cc", 'reg_constant': 0.000, 'max_grad_norm': 10.0, 'num_epochs': 100, # number of epochs over dataset for training 'start_epoch': 1, # epoch to start 'reset_global_step': True,